thomascorcoral / ml_sound

Wildlife Identification Learning Tools - ML Sound

Wildlife Identification Learning Tools - ML Sound

Audio Recording Classification by Machine Learning

The windows installer is now available here

The documentation is available here : Documentation

Installation

WINDOWS :

• Download the installer here
• Install the program
• Start the program with admin rights
• If you have a problem with .mp3 files, try to add ffmpeg path to PATH environment variable

LINUX :

• Clone the project
• Install the dependencies requirements.txt
• start the main.py file with a python 3 interpreter

Train

To use .mp3 please follow those steps:

WINDOWS :

• Fill in the path to the absolute path /ffmpeg folder in the PATH environment variable
• Unzip the ffmpeg.7z file in the ffmpeg

LINUX :

• Install ffmpeg throught sudo apt install ffmpeg

main.py Start the UI application, you can use your own data to train your model. Select your data path by clicking the button. Then generate the CSV file or use your own (be careful to respect the rules write in the help section). Then you will need to format your data (This step can take a long time to complete). When your data is format, you can train your model.

Test

main.py The UI app allows you once your data is prepared and your model is trained to classify audio recording. Click on the button to choose your your test file and then you can do a prediction with your pretrain model.

GUI application

In the main directory you'll find main.py which will allow you to launch the application that will allow you to prepare your data and train the model. It is also possible to view (audio, spectrogram and mfcc) the file you want to classify.

Misc

Organisation

In order to organize our project we first created this github as well as a Trello. A website followed in order to be able to access easily and at any time a lot of information (see below).

Usage

To be able to use the application, install python 3.8 and the dependencies present in the file requirements.txt. Once this is done, launch the main.py file, for more information click on the "Help" button at the bottom right of the application

Application

The application developed in python uses the tkinter library to perform the following display:

Website

Visit it

This website contains :

• A training section to learn the basics to contribute to the project
• Different resources + meeting minutes
• Project documentation

Dependencies

requirements.txt

librosa~=0.8.0
matplotlib~=3.3.3
numpy~=1.19.4
scipy~=1.5.4
tensorflow~=2.4.0
scikit-learn~=0.23.2
Keras~=2.4.3
pandas~=1.1.5
pydub~=0.24.1
future~=0.18.2
pygame~=2.0.1
PyAudio~=0.2.11
noisereduce~=1.1.0

References

@misc{smales_2020, title={Sound Classification using Deep Learning}, url={https://mikesmales.medium.com/sound-classification-using-deep-learning-8bc2aa1990b7}, journal={Medium}, publisher={Medium}, author={Smales, Mike}, year={2020}, month={Nov}}

@inproceedings{batdetect18, title={Bat Detective - Deep Learning Tools for Bat Acoustic Signal Detection}, author={Mac Aodha, Oisin and Gibb, Rory and Barlow, Kate and Browning, Ella and Firman, Michael and   Freeman, Robin and Harder, Briana and Kinsey, Libby and Mead, Gary and Newson, Stuart and Pandourski, Ivan and Parsons, Stuart and Russ, Jon and Szodoray-Paradi, Abigel and Szodoray-Paradi, Farkas and Tilova, Elena and Girolami, Mark and Brostow, Gabriel and E. Jones, Kate.}, journal={PLOS Computational Biology}, year={2018}}

@misc{gavali_mhetre_patil_bamane_buva_2019, title={Bird Species Identification using Deep Learning}, url={https://www.ijert.org/bird-species-identification-using-deep-learning#:~:text=200 [CUB-200-2011]) shows that algorithm achieves,between 80% and 90%.&text=Birds help us to detect,the environmental changes [2].}, journal={International Journal of Engineering Research & Technology}, publisher={IJERT-International Journal of Engineering Research & Technology}, author={Gavali, Prof. Pralhad and Mhetre, Ms. Prachi Abhijeet and Patil, Ms. Neha Chandrakhant and Bamane, Ms. Nikita Suresh and Buva, Ms. Harshal Dipak}, year={2019}, month={Apr}}

@misc{kortas_2020, title={Sound-Based Bird Classification}, url={https://towardsdatascience.com/sound-based-bird-classification-965d0ecacb2b}, journal={Medium}, publisher={Towards Data Science}, author={Kortas, Magdalena}, year={2020}, month={Jan}}

@misc{salamon_jacoby_bello_new york_new york, title={UrbanSound8K}, url={https://urbansounddataset.weebly.com/urbansound8k.html}, journal={Urban Sound Datasets}, author={Salamon, Justin and Jacoby, Christopher and Bello, Juan Pablo and New York, MARL and New York, CUSP}}

@misc{bushaev_2018, title={Adam - latest trends in deep learning optimization.}, url={https://towardsdatascience.com/adam-latest-trends-in-deep-learning-optimization-6be9a291375c#:~:text=Adam [1] is an adaptive,for training deep neural networks.&text=The algorithms leverages the,learning rates for each parameter}, journal={Medium}, publisher={Towards Data Science}, author={Bushaev, Vitaly}, year={2018}, month={Oct}}